To systematically identify genes related to invasion a three-dimensional multicellular matrix invasion assay was used to classify human tumor cell lines as stromal invasion positive or stromal invasion negative. Cells from two of the primary cell types of the stromal compartment [endothelial cells (HMVEC) and myofibroblasts (HDF)] were assayed for invasion into tumor cell clusters (breast carcinoma, ovarian carcinoma, prostate carcinoma, lung carcinoma, and melanoma). Four tumor cell lines (MDA-MB231, SKOV-3, A375, and MEL624) scored invasion positive, and four tumor cell lines (LNCaP, DU145, PC3, and A549) scored invasion negative. Serial analysis of gene expression (SAGE) libraries generated from the tumor cell lines were analyzed by GeneSpring Hierarchical clustering, t test, and chi(2) test. Clusters emerged that reflected the behavior in the cell culture assay. Of the 47 most highly differentially expressed genes, 30 were selected for confirmation by real-time PCR, and 9 had good correlation with normalized serial analysis of gene expression tag counts. The strongest correlations were for bone marrow stromal antigen 2, stathmin-like 3, tumor necrosis factor receptor superfamily member 5, and hepatocyte growth factor tyrosine kinase substrate. In situ hybridization of metastatic and nonmetastatic ovarian cancer demonstrated selective expression of bone marrow stromal antigen 2 and tumor necrosis factor receptor superfamily member 5 in the metastatic disease. This combination approach appears to be a powerful tool for identifying genes that may be useful as diagnostic markers and/or as therapeutic targets for invasive solid tumors.